Video super resolution is a process for recovering a high resolution (HR) video from a low resolution (LR) video. Video super resolution improves quality and retrieves information from low-resolution videos that have been used in mobile/smart phone cameras. Among some typical video super resolution systems, a deep CNN is used.
The accuracy of a conventional video SR system is limited since temporal motion information is not used well. Conventional video SR methods only use the motion in the LR domain.
An issue in constructing a CNN-based video super resolution system is related to the use of temporal information to align frames. Most conventional methods receive an optical flow in LR only, and apply image warping for motion compensation. The performance is limited because only LR motion is considered.